76 points by adunk 10 hours ago | 11 comments
mintflow 18 minutes ago
This is cool, let aside the token usage, perhaps it can help analyze tcp throughput by redirect wire shark/to dump result
ForHackernews 17 minutes ago
>Fun? Oh yeah!

I think this author and I have different definitions of fun.

ShinyLeftPad 2 hours ago
How quickly claude responds when it acts like a user space LLM chatbot?
twoodfin 3 hours ago
Modulo Anthropic messing with the model for load mitigation, I wonder how stable this result is.

1,000 pings, how many correctly ponged?

fouc 4 hours ago
think about how much faster it would've been with a small local model!
6 hours ago
ValdikSS 7 hours ago
That's why LLM will eventually be used only for initial interaction between the user in their language, to prepare the data to a specialized model.

Imagine face recognition to work like a text chat, where the PC gets the frame from the camera and writes in the chat: "Who's that? Here's the RGB888 image in hex: ...".

FeepingCreature 3 hours ago
That's actually how vision language models already work, pretty much.
wongarsu 33 minutes ago
And there's a reason nobody uses them for face recognition

Vision language models are an incredible achievement in the generality and usability. But they pay a hefty price in fidelity and speed

stingraycharles 3 hours ago
Huh? The images are tokenized in the same way language is and it’s just fed into one single model. Not multiple smaller expert models.

Image gets rasterized into smaller pieces (eg 4x4 pixels) and each of those is assigned a token, similarly how text is broken up into tokens. And the whole thing is fed into a single model.

FeepingCreature 2 hours ago
Yes I'm saying

> Imagine face recognition to work like a text chat, where the PC gets the frame from the camera and writes in the chat: "Who's that? Here's the RGB888 image in hex: ...".

that's p much how it works.

stingraycharles 47 minutes ago
But that isn’t a specialized model like the grandparent claimed, but rather a single, multi-modal model.
stingraycharles 3 hours ago
Do you know that MoE is a thing?
jampekka 3 hours ago
The experts in MoEs aren't specialized in any meaningful task sense. From level of what we would think as tasks MoEs are selected essentially arbitrarily per token and per block.
stingraycharles 2 hours ago
It’s unsupervised, yes, but “unspecialized in any meaningful task sense” is incorrect, that’s the whole point. It’s just not in the sense of “this is a legal expert, this is a software developer”.
bot403 1 hour ago
Now do the equivalent of just in time compilation. Claude sees that we need to respond to a lot of pings and writes a program to compute it instead of thinking about each one.
westurner 6 hours ago
Wouldn't this be faster with an agent skill that has code?

/skill-creator [or /create-skill] Write an agent skill with code script(s) that use an existing user space IP library that works with your agent runtime, to [...]

ComposioHQ/awesome-claude-skills: https://github.com/ComposioHQ/awesome-claude-skills

anthopics/skills//skill-creator/SKILL.md: https://github.com/anthropics/skills/blob/main/skills/skill-...

/.agents/skills/skill-name/SKILL.md, scripts/{script_name.py,__init__.py}

https://agentskills.io/what-are-skills

trollbridge 6 hours ago
Well, yeah, of course it would be.

Even faster would just to be use code in the first place!

brcmthrowaway 7 hours ago
Next up: Claude replacement to handle simdjson processing.
jeremyjh 5 hours ago
Perhaps one day, all network services will be provided by LLMs natively. Truly, that would be a day in the future.
pastage 2 hours ago
You could read about that in 1992 "A Fire Upon the Deep" by Vernor Vinge. There is prompt injection in communication, in the book certain protocols for information communication can not be deterministic so if someone is too smart you get hacked.
lionkor 1 hour ago
"Perhaps" doing enough lifting to participate in a bodybuilder contest, in that sentence
vrighter 4 hours ago
why? We already have more efficient specialized hardware.
codezero 5 hours ago
I mean, we did decades of JavaScript, so... I mean... anything is possible, right? :)